xref: /xnu-8020.101.4/bsd/miscfs/nullfs/null_subr.c (revision e7776783b89a353188416a9a346c6cdb4928faad)
1 /*
2  * Copyright (c) 2016 Apple Inc. All rights reserved.
3  *
4  * @APPLE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. Please obtain a copy of the License at
10  * http://www.opensource.apple.com/apsl/ and read it before using this
11  * file.
12  *
13  * The Original Code and all software distributed under the License are
14  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
15  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
16  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
18  * Please see the License for the specific language governing rights and
19  * limitations under the License.
20  *
21  * @APPLE_LICENSE_HEADER_END@
22  */
23 
24 /*-
25  * Portions Copyright (c) 1992, 1993
26  *  The Regents of the University of California.  All rights reserved.
27  *
28  * This code is derived from software donated to Berkeley by
29  * Jan-Simon Pendry.
30  *
31  * Redistribution and use in source and binary forms, with or without
32  * modification, are permitted provided that the following conditions
33  * are met:
34  * 1. Redistributions of source code must retain the above copyright
35  *    notice, this list of conditions and the following disclaimer.
36  * 2. Redistributions in binary form must reproduce the above copyright
37  *    notice, this list of conditions and the following disclaimer in the
38  *    documentation and/or other materials provided with the distribution.
39  * 4. Neither the name of the University nor the names of its contributors
40  *    may be used to endorse or promote products derived from this software
41  *    without specific prior written permission.
42  *
43  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
44  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
47  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53  * SUCH DAMAGE.
54  *
55  *  @(#)null_subr.c 8.7 (Berkeley) 5/14/95
56  *
57  * $FreeBSD$
58  */
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/lock.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/proc.h>
66 #include <sys/vnode.h>
67 
68 #include "nullfs.h"
69 
70 /*
71  * Null layer cache:
72  * Each cache entry holds a reference to the lower vnode
73  * along with a pointer to the alias vnode.  When an
74  * entry is added the lower vnode is VREF'd.  When the
75  * alias is removed the lower vnode is vrele'd.
76  */
77 
78 #define NULL_HASH_SIZE (desiredvnodes / 10)
79 
80 /* osx doesn't really have the functionality freebsd uses here..gonna try this
81  * hacked hash...*/
82 #define NULL_NHASH(vp) (&null_node_hashtbl[((((uintptr_t)vp) >> vnsz2log) + (uintptr_t)vnode_mount(vp)) & null_hash_mask])
83 
84 static LIST_HEAD(null_node_hashhead, null_node) * null_node_hashtbl;
85 static LCK_GRP_DECLARE(null_hashlck_grp, "com.apple.filesystems.nullfs");
86 static LCK_MTX_DECLARE(null_hashmtx, &null_hashlck_grp);
87 static u_long null_hash_mask;
88 
89 /* os x doesn't have hashes built into vnode. gonna try doing what freebsd does
90  *  anyway
91  *  Don't want to create a dependency on vnode_internal.h and the real struct
92  *  vnode.
93  *  9 is an eyeball of the log 2 size of vnode */
94 static int vnsz2log = 9;
95 
96 static int null_hashins(struct mount *, struct null_node *, struct vnode **);
97 
98 void
nullfs_init_lck(lck_mtx_t * lck)99 nullfs_init_lck(lck_mtx_t * lck)
100 {
101 	lck_mtx_init(lck, &null_hashlck_grp, LCK_ATTR_NULL);
102 }
103 
104 void
nullfs_destroy_lck(lck_mtx_t * lck)105 nullfs_destroy_lck(lck_mtx_t * lck)
106 {
107 	lck_mtx_destroy(lck, &null_hashlck_grp);
108 }
109 
110 /*
111  * Initialise cache headers
112  */
113 int
nullfs_init(__unused struct vfsconf * vfsp)114 nullfs_init(__unused struct vfsconf * vfsp)
115 {
116 	NULLFSDEBUG("%s\n", __FUNCTION__);
117 	null_node_hashtbl = hashinit(NULL_HASH_SIZE, M_TEMP, &null_hash_mask);
118 	NULLFSDEBUG("%s finished\n", __FUNCTION__);
119 	return 0;
120 }
121 
122 int
nullfs_uninit(void)123 nullfs_uninit(void)
124 {
125 	/* This gets called when the fs is uninstalled, there wasn't an exact
126 	 * equivalent in vfsops */
127 	hashdestroy(null_node_hashtbl, M_TEMP, null_hash_mask);
128 	return 0;
129 }
130 
131 /*
132  * Find the nullfs vnode mapped to lowervp. Return it in *vpp with an iocount if found.
133  * Return 0 on success. On failure *vpp will be null and a non-zero error code will be returned.
134  */
135 int
null_hashget(struct mount * mp,struct vnode * lowervp,struct vnode ** vpp)136 null_hashget(struct mount * mp, struct vnode * lowervp, struct vnode ** vpp)
137 {
138 	struct null_node_hashhead * hd = NULL;
139 	struct null_node * a = NULL;
140 	struct vnode * vp = NULL;
141 	uint32_t vp_vid = 0;
142 	int error = ENOENT;
143 
144 	/*
145 	 * Find hash base, and then search the (two-way) linked
146 	 * list looking for a null_node structure which is referencing
147 	 * the lower vnode. We only give up our reference at reclaim so
148 	 * just check whether the lowervp has gotten pulled from under us
149 	 */
150 	hd = NULL_NHASH(lowervp);
151 	// In the future we should consider using a per bucket lock
152 	lck_mtx_lock(&null_hashmtx);
153 	LIST_FOREACH(a, hd, null_hash)
154 	{
155 		if (a->null_lowervp == lowervp && vnode_mount(NULLTOV(a)) == mp) {
156 			vp = NULLTOV(a);
157 			if (a->null_lowervid != vnode_vid(lowervp)) {
158 				/*lowervp has reved */
159 				error = EIO;
160 				vp = NULL;
161 			} else {
162 				vp_vid = a->null_myvid;
163 			}
164 			// In the case of a succesful look-up we should consider moving the object to the top of the head
165 			break;
166 		}
167 	}
168 	lck_mtx_unlock(&null_hashmtx);
169 	if (vp != NULL) {
170 		error = vnode_getwithvid(vp, vp_vid);
171 		if (error == 0) {
172 			*vpp = vp;
173 		}
174 	}
175 	return error;
176 }
177 
178 /*
179  * Act like null_hashget, but add passed null_node to hash if no existing
180  * node found.
181  */
182 static int
null_hashins(struct mount * mp,struct null_node * xp,struct vnode ** vpp)183 null_hashins(struct mount * mp, struct null_node * xp, struct vnode ** vpp)
184 {
185 	struct null_node_hashhead * hd = NULL;
186 	struct null_node * oxp = NULL;
187 	struct vnode * ovp = NULL;
188 	uint32_t oxp_vid = 0;
189 	int error = 0;
190 
191 	hd = NULL_NHASH(xp->null_lowervp);
192 	lck_mtx_lock(&null_hashmtx);
193 	LIST_FOREACH(oxp, hd, null_hash)
194 	{
195 		if (oxp->null_lowervp == xp->null_lowervp && vnode_mount(NULLTOV(oxp)) == mp) {
196 			/*
197 			 * See null_hashget for a description of this
198 			 * operation.
199 			 */
200 			ovp = NULLTOV(oxp);
201 			if (oxp->null_lowervid != vnode_vid(oxp->null_lowervp)) {
202 				/*vp doesn't exist so return null (not sure we are actually gonna catch
203 				 *  recycle right now
204 				 *  This is an exceptional case right now, it suggests the vnode we are
205 				 *  trying to add has been recycled
206 				 *  don't add it.*/
207 				error = EIO;
208 				ovp = NULL;
209 			} else {
210 				oxp_vid = oxp->null_myvid;
211 			}
212 			goto end;
213 		}
214 	}
215 	/* if it wasn't in the hash map then the vnode pointed to by xp already has a
216 	 * iocount so don't bother */
217 	LIST_INSERT_HEAD(hd, xp, null_hash);
218 	xp->null_flags |= NULL_FLAG_HASHED;
219 end:
220 	lck_mtx_unlock(&null_hashmtx);
221 	if (ovp != NULL) {
222 		/* if we found something in the hash map then grab an iocount */
223 		error = vnode_getwithvid(ovp, oxp_vid);
224 		if (error == 0) {
225 			*vpp = ovp;
226 		}
227 	}
228 	return error;
229 }
230 
231 /*
232  * Remove node from hash.
233  */
234 void
null_hashrem(struct null_node * xp)235 null_hashrem(struct null_node * xp)
236 {
237 	lck_mtx_lock(&null_hashmtx);
238 	LIST_REMOVE(xp, null_hash);
239 	lck_mtx_unlock(&null_hashmtx);
240 }
241 
242 static struct null_node *
null_nodecreate(struct vnode * lowervp)243 null_nodecreate(struct vnode * lowervp)
244 {
245 	struct null_node * xp;
246 
247 	xp = kalloc_type(struct null_node, Z_WAITOK | Z_ZERO | Z_NOFAIL);
248 	if (lowervp) {
249 		xp->null_lowervp  = lowervp;
250 		xp->null_lowervid = vnode_vid(lowervp);
251 	}
252 	return xp;
253 }
254 
255 /* assumption is that vnode has iocount on it after vnode create */
256 int
null_getnewvnode(struct mount * mp,struct vnode * lowervp,struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,int root)257 null_getnewvnode(
258 	struct mount * mp, struct vnode * lowervp, struct vnode * dvp, struct vnode ** vpp, struct componentname * cnp, int root)
259 {
260 	struct vnode_fsparam vnfs_param;
261 	int error             = 0;
262 	enum vtype type       = VDIR;
263 	struct null_node * xp = null_nodecreate(lowervp);
264 
265 	if (xp == NULL) {
266 		return ENOMEM;
267 	}
268 
269 	if (lowervp) {
270 		type = vnode_vtype(lowervp);
271 	}
272 
273 	vnfs_param.vnfs_mp         = mp;
274 	vnfs_param.vnfs_vtype      = type;
275 	vnfs_param.vnfs_str        = "nullfs";
276 	vnfs_param.vnfs_dvp        = dvp;
277 	vnfs_param.vnfs_fsnode     = (void *)xp;
278 	vnfs_param.vnfs_vops       = nullfs_vnodeop_p;
279 	vnfs_param.vnfs_markroot   = root;
280 	vnfs_param.vnfs_marksystem = 0;
281 	vnfs_param.vnfs_rdev       = 0;
282 	vnfs_param.vnfs_filesize   = 0; // set this to 0 since we should only be shadowing non-regular files
283 	vnfs_param.vnfs_cnp        = cnp;
284 	vnfs_param.vnfs_flags      = VNFS_ADDFSREF;
285 
286 	error = vnode_create(VNCREATE_FLAVOR, VCREATESIZE, &vnfs_param, vpp);
287 	if (error == 0) {
288 		xp->null_vnode = *vpp;
289 		xp->null_myvid = vnode_vid(*vpp);
290 		vnode_settag(*vpp, VT_NULL);
291 	} else {
292 		kfree_type(struct null_node, xp);
293 	}
294 	return error;
295 }
296 
297 /*
298  * Make a new or get existing nullfs node.
299  * Vp is the alias vnode, lowervp is the lower vnode.
300  *
301  * lowervp is assumed to have an iocount on it from the caller
302  */
303 int
null_nodeget(struct mount * mp,struct vnode * lowervp,struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,int root)304 null_nodeget(
305 	struct mount * mp, struct vnode * lowervp, struct vnode * dvp, struct vnode ** vpp, struct componentname * cnp, int root)
306 {
307 	struct vnode * vp;
308 	int error;
309 
310 	/* Lookup the hash firstly. */
311 	error = null_hashget(mp, lowervp, vpp);
312 	/* ENOENT means it wasn't found, EIO is a failure we should bail from, 0 is it
313 	 * was found */
314 	if (error != ENOENT) {
315 		/* null_hashget checked the vid, so if we got something here its legit to
316 		 * the best of our knowledge*/
317 		/* if we found something then there is an iocount on vpp,
318 		 *  if we didn't find something then vpp shouldn't be used by the caller */
319 		return error;
320 	}
321 
322 	/*
323 	 * We do not serialize vnode creation, instead we will check for
324 	 * duplicates later, when adding new vnode to hash.
325 	 */
326 	error = vnode_ref(lowervp); // take a ref on lowervp so we let the system know we care about it
327 	if (error) {
328 		// Failed to get a reference on the lower vp so bail. Lowervp may be gone already.
329 		return error;
330 	}
331 
332 	error = null_getnewvnode(mp, lowervp, dvp, &vp, cnp, root);
333 
334 	if (error) {
335 		vnode_rele(lowervp);
336 		return error;
337 	}
338 
339 	/*
340 	 * Atomically insert our new node into the hash or vget existing
341 	 * if someone else has beaten us to it.
342 	 */
343 	error = null_hashins(mp, VTONULL(vp), vpp);
344 	if (error || *vpp != NULL) {
345 		/* recycle will call reclaim which will get rid of the internals */
346 		vnode_recycle(vp);
347 		vnode_put(vp);
348 		/* if we found vpp, then null_hashins put an iocount on it */
349 		return error;
350 	}
351 
352 	/* vp has an iocount from null_getnewvnode */
353 	*vpp = vp;
354 
355 	return 0;
356 }
357